← TrustMyChoice
Founding evidence program · JavaScript + agent skill

Turn real service calls into useful public evidence.

Check a service before an agent uses it, then contribute success, reliability, latency, and optional price metrics afterward. The observer never sends the service input, output, credential, or error body.

Before

Check the exact version and task

The public lookup exposes evidence, uncertainty, recency, and correction state. Insufficient evidence stays explicit.

During

Measure only the service call

The observer wraps the existing call and measures latency without proxying the request through ScrubMyText.

After

Submit bounded metrics

A stable observation key makes retries safe. The primary service result remains local and reporting failure does not change it.

Public threshold

Three independent sources

A card becomes public only after three observations from three distinct trial or paid accounts match the exact version and task.

JavaScript quickstart

Save the zero-dependency module locally, then wrap only the external service call you want to measure.

curl -o trustmychoice-observer.mjs https://scrubmytext.com/sdk/trustmychoice-observer.mjs
import { createTrustMyChoiceObserver } from './trustmychoice-observer.mjs';

const observer = createTrustMyChoiceObserver({
  apiKey: process.env.SCRUBMYTEXT_API_KEY,
  target: {
    type: 'mcp_server',
    providerDomain: 'api.example.com',
    canonicalEndpoint: 'https://api.example.com/mcp',
    version: '2.1.0'
  },
  task: { category: 'invoice-extraction', maxLatencyMs: 2000 },
  onReportError: error => console.warn(error.code)
});

const result = await observer.observe({
  observationKey: job.id,
  execute: () => invoiceService.extract(job.document),
  classify: ({ value, error, latencyMs }) => ({
    outcomeSuccess: !error && value.fieldsComplete,
    reliability: !error,
    latencyMs
  })
});
Agent quickstart

Give an agent the complete check-and-report workflow

The portable skill instructs an agent to consult contextual evidence before service selection and to submit only opt-in structured metrics afterward. Place the downloaded folder in your agent’s skills directory and connect the ScrubMyText MCP server at https://api.scrubmytext.com/mcp.

Read the skill source before installing →

Help create the first five useful cards

We are starting with web research, browser automation, document extraction, code execution, and structured-data enrichment. Founding participants receive hands-on setup during the seven-day trial. Participation is rewarded with support—not for producing a positive outcome.

Join the founding program

Card required. Unless canceled before the trial ends, Starter renews automatically at $15/month. No automatic overages.

Neutral evidence only

Do not create duplicate accounts, manufacture observations, reward positive results, or report a failure caused by unrelated caller code as a provider failure. An empty card is better than a misleading one.